prismnote 0.4.4

Modern, open-source Jupyter-compatible data science notebook with Rust performance

◆ PrismNote

The fast, local-first data-science notebook — with a warehouse-grade Data Explorer built in.

Python + SQL · a BigQuery/Snowflake-style Data Explorer · no-code charts · local AI · jobs · git — all on your machine.

⭐ If PrismNote is useful to you, please star the repo — it genuinely helps the project grow and reach more data scientists.

---

Why PrismNote?

Jupyter is great for code, but exploring data still means writing df.head(), df.describe(), and one-off matplotlib calls. PrismNote makes data exploration a first-class surface: open any DataFrame — or a Parquet/CSV/Iceberg file — in a fast, scrollable grid with per-column statistics, distributions, and lineage, then build charts with no code. It's local-first (your data never leaves your machine), Jupyter-compatible (.ipynb in/out), and AI-native (local via Ollama, or Claude/OpenAI).

• 🔍 Data Explorer first — the headline feature, ranked above the notebook itself.
• 🔒 Local & private — runs on your machine; local AI via Ollama, no account required.
• 🦀 Fast — a Rust + Axum engine driving a persistent Python kernel.
• 🔗 Open formats — Parquet, CSV, JSON, Arrow, Apache Iceberg via DuckDB.
• 🧩 Batteries included — kernel, SQL, charts, AI, jobs, git, deploy, search, terminal.

---

Quickstart

Install with pip or uv or brew or curl — pick one:

# ── pip ──
pip install prismnote && prismnote

— OR —

# ── uv ──  run instantly (no install), or install as a tool
uvx prismnote
#   or:  uv tool install prismnote && prismnote

— OR —

# ── Homebrew (macOS / Linux) ──
brew tap Mullassery/prismnote
brew install prismnote && prismnote

— OR —

# ── curl ──  no Python required (installs the binary to /usr/local/bin)
curl -fsSL https://raw.githubusercontent.com/Mullassery/prismnote/main/install.sh | bash
prismnote

Then open http://localhost:8000 and click Open Data Explorer (⌘E) or New Notebook.

No Docker required. PrismNote runs as a single native binary — Docker is not needed to install or run it. (Docker is only used by the optional Deploy to Cloud artifacts, and by maintainers who cross-build release binaries.)

The pip/uv package is a thin launcher: on first run it downloads the prebuilt server binary for your platform from the matching GitHub Release (vX.Y.Z). If a release binary isn't published for your platform yet, use From source below.

From source

Requires Rust (stable), Node 18+, and Python 3.8+ on PATH:

# kernel runtime deps (ipykernel required; the rest enable rich output, %sql, charts)
pip install ipykernel pandas matplotlib rich duckdb

cargo run                                   # backend  → http://localhost:8000
cd frontend && npm install && npm run dev   # frontend → http://localhost:5173

Open http://localhost:5173.

---

Highlights

🔍 Data Explorer — better than a df.describe()

Open a live DataFrame, a file, or a DuckDB query in a warehouse-style explorer:

• Virtualized grid that scrolls millions of rows (server-side paging/sort/filter).
• Tabs: Preview · Schema (types + null %) · Statistics · Metadata · Lineage.
• Per-column profiling — histograms, top values, and a full describe() table (count, nulls, distinct, mean, std, min/quartiles/max, sum, skew, kurtosis).
• Nested types — struct and array columns are detected and rendered.
• Open formats via DuckDB — Parquet, CSV, JSON, Arrow, Apache Iceberg; or any DuckDB SQL.
• Lineage — provenance (variable / file / query), upstream sources, and the notebook cells that define and use the dataset.
• Reproducible — "Copy as code" / "Insert as cell" emits runnable pandas; export CSV.
• Collapsible, with +/- zoom; double-click a data file in the file browser to open it here.

📊 Visualization Pane

• Plot gallery — every figure (matplotlib/plotly) collected with zoom, pan, filmstrip, and PNG + vector SVG export.
• No-code chart builder (Looker-style) — drag dimensions/measures, pick a chart (bar/line/area/scatter/heatmap/pie), aggregate, and render via Vega-Lite — then "Copy as code" to get the equivalent Altair.

🤖 AI assistance

• Choose your provider in Settings → AI: Ollama (local, free, private), Claude, or OpenAI — with model pickers and live connection status.
• In-cell ⌘K edit (diff accept/reject), Fix with AI on errors, Explain.
• Inline autocomplete (ghost text) and a Plan/Act agent that's aware of your workspace files and the open dataset.

📓 Notebook & execution

• Persistent shared kernel — state carries across cells; interrupt & restart.
• Magics: %python, %sql (in-process DuckDB), %sh / !cmd, %md.
• Rich output, live-streamed over WebSocket; input widgets that re-run cells.
• Auto-formatting — code is pretty-printed with Black on paste and on ⇧⌥F.
• Friendly errors — plain-language explanations + in-editor markers.

🗄️ Data & SQL

• Connect to SQLite, DuckDB, PostgreSQL, MySQL, and 8 cloud warehouses (Snowflake, BigQuery, Redshift, Databricks, Athena, Trino, Presto, Synapse).
• Queries run through the kernel using permissively-licensed OSS drivers you install — nothing proprietary vendored. See CONNECTORS.md.

⚙️ Workflow

• Jobs — run a whole notebook on a schedule (manual / interval / daily) with run history; Airflow trigger + generated DAG.
• Source control — init / clone / commit / push / pull / status from the UI.
• Cloud deploy — generates Dockerfile, docker-compose.yml, k8s.yaml, fly.toml.

---

How PrismNote compares

PrismNote focuses on data exploration, local AI, and a single-binary local-first experience. JupyterLab, Apache Zeppelin, and PyCharm are mature, broader tools — here's an honest side-by-side.

✅ built-in · ⚠️ via extension / partial / paid · ❌ not available · 🔜 on the roadmap

Feature	PrismNote	JupyterLab	Apache Zeppelin	PyCharm (Pro)
Type	Notebook (web)	Notebook (web)	Notebook (web)	Desktop IDE
Engine / runtime	Rust + Python kernel	Python (Jupyter)	JVM + interpreters	JetBrains (JVM) + Python
.ipynb format (native)	✅	✅	⚠️ (own format)	✅
Built-in Data Explorer (grid, sort/filter)	✅	⚠️	⚠️ (basic)	✅ (DataFrame viewer)
Column profiling, describe() & lineage	✅	❌	❌	⚠️ (column stats)
Open-format explorer (Parquet/CSV/Iceberg)	✅ (DuckDB)	⚠️ (code)	⚠️ (code)	⚠️ (DB tools/CSV)
No-code chart builder	✅ (Vega-Lite)	❌ (code)	✅ (on SQL)	⚠️ (viewer charts)
In-process SQL (DuckDB) + magics	✅	⚠️	✅	⚠️ (DB tools)
AI: in-cell edit / fix / explain / agent	✅ (Ollama/Claude/OpenAI)	⚠️ (Jupyter AI)	❌	⚠️ (AI Assistant, paid)
AI inline autocomplete	✅	⚠️	❌	✅
Code auto-format (Black)	✅	⚠️	❌	✅
Variable explorer	✅	⚠️	❌	✅
Scheduled jobs (built-in)	✅	❌ (external)	✅ (cron)	❌
Git integration (UI)	✅	⚠️ (jupyterlab-git)	⚠️	✅ (excellent)
Cloud deploy artifacts (Docker/k8s/Fly)	✅	❌	❌	⚠️ (plugins)
Real-time collaboration	🔜	✅	⚠️	✅ (Code With Me)
Multi-language interpreters	⚠️ (Python/SQL/shell)	✅ (many kernels)	✅ (many)	⚠️ (Python focus)
Install	pip / uv / single binary	pip / conda	download + JVM	IDE download
License	MIT	BSD-3	Apache-2.0	Proprietary (Pro)

In short: choose JupyterLab for its vast kernel/extension ecosystem and real-time collaboration, Zeppelin for JVM/Spark-centric multi-interpreter workloads, PyCharm for a full desktop IDE with deep refactoring and DB tools (paid Pro), and PrismNote when you want fast, no-setup data exploration + charts + local AI in one free, open-source local binary. See also ZEPPELIN_COMPARISON.md and NOTEBOOK_COMPARISON_MATRIX.md.

---

Configure AI (optional)

Local — Ollama (recommended for privacy/cost):

# install from https://ollama.com, then pull a coding model
ollama pull qwen2.5-coder
# the browser UI talks to Ollama directly, so allow the origin once:
OLLAMA_ORIGINS=http://localhost:5173 ollama serve

Claude / OpenAI — set it in Settings → AI Provider (keys are stored locally in ~/.prismnote/ai_config.json and never leave your machine), or via env:

export PRISMNOTE_AI_PROVIDER=claude      # or openai / ollama
export ANTHROPIC_API_KEY=...             # or OPENAI_API_KEY

---

Keyboard shortcuts

Shortcut	Action
⌘E	Open Data Explorer
⌘N / ⌘O / ⌘S	New / Open / Save notebook
⌘K	Global search (in a focused cell, ⌘K = AI edit)
⇧⌘P	Command palette
⌘,	Settings
⌘⇧↵ / ⇧↵	Run all / run focused cell
⇧⌥F	Format cell (Black)

(⌘/Ctrl depending on platform.)

---

Architecture

┌──────────────────────────────┐        ┌─────────────────────────────┐
│  React + TypeScript (Vite)   │  HTTP  │      Rust backend (Axum)    │
│  Monaco · Tailwind · zustand │ ─────▶ │  REST + WebSocket           │
│  explorer · charts · AI      │ ◀───── │  explore · jobs · git · …   │
└──────────────────────────────┘   WS   └──────────────┬──────────────┘
                                                        │ stdin/stdout (JSON)
                                                ┌───────▼────────┐
                                                │ Persistent      │
                                                │ Python kernel   │
                                                │ (shared globals)│
                                                └─────────────────┘

The backend spawns one long-lived python process and speaks a line-framed JSON protocol; outputs are Jupyter-style MIME bundles, streamed live. The Data Explorer pushes profiling/paging/aggregation into the kernel where the data already lives.

---

Deploy

Open Deploy to Cloud to copy/download the generated artifacts, then:

docker compose up -d            # Docker
kubectl apply -f k8s.yaml       # Kubernetes
fly launch --copy-config --now  # Fly.io

---

Project layout

crates/server/   Rust backend (api, kernel, explore, jobs, db, ws, deploy, git…)
frontend/        React app (components, hooks, api clients)
python/          PyPI launcher package (prismnote)
docs/            screenshots & comparison docs

Further reading: CONNECTORS.md · ZEPPELIN_COMPARISON.md · DATABRICKS_COMPARISON.md · NOTEBOOK_COMPARISON_MATRIX.md

---

Roadmap

• Prebuilt release binaries for all platforms (so pip install runs out of the box).
• Distributed compute (Spark) and a catalog/data browser.
• Real-time collaboration (live cursors / co-editing).
• Notebook parameters and multi-notebook job composition.
• Reactive (dependency-aware) cell execution.

---

Contributing

⭐ The easiest way to contribute is to star the repo — it boosts visibility and helps others discover PrismNote.

Code contributions are welcome too! Please open an issue to discuss substantial changes first.

git clone https://github.com/Mullassery/prismnote.git
cd prismnote
cargo run                                   # backend
cd frontend && npm install && npm run dev   # frontend

Run cargo check and npm run build before opening a PR. See CONTRIBUTING.md for details.

---

License

MIT © Georgi Mammen Mullassery